Embedded XML Web Services

1
Embedded XML Web Services

• Realizing a Seamless and Interoperable World
  through Invisible Computing
• Johannes Helander
• Researcher
• Microsoft Research
• March 2, 2005

2
Research Agenda

• Make deeply embedded devices interoperable with
  all classes of computers
• Extend XML Web Services to new areas
• Implement prototype software
• Provide Microsoft presence into new area
• Add value to PC and services such as ASP.NET

3
The Goal Invisible Computing

• Make everyday objects better by adding
  computation and communication
• 5 microcontrollers (32-bit single chip computer)
• The computers do their job out of sight
• Natural user interface not screen and mouse
• Incremental deployment, no setup hassles
• Operate autonomously added value from services
• Devices communicate with each other, PCs, Web

4
Sample Applications

• Home appliances, security, lighting
• Sensor networks
• Medical electronic devices
• Wearable computers
• Robotics, industrial control, national
  infrastructure

• Wireless communication gadgets
• Audio net
• Disaggregated PC, smart I/O cards
• Toys

5
Outline

• Invisible Computing Devices
• Why Web Services?
• Our approach
• Table driven serialization
• Compressed SOAP
• Trust and secure discovery
• Distributed real-time
• Componentized RTOS
• Real-time C
• Result
• Call to action

6

VCR
XP Embedded

Pacemaker

• Interoperability
• Security
• Data analysis

Watch

• Power
• Bandwidth
• Processing
• Routing
• Security
• Real-Time
• Non-graphical UI
• Zero-configuration

An Invisible Computing Scenario
7
What are Web Services?

• The general-purpose solution to communication, in
  XML
• XML is more suitable to computer processing than
  HTML. Common language for representing any data.
• Convergence of EDI, RPC, MSMQ, app specific
  protocols and formats

• All about interoperation. Allows partial
  understanding
• Common protocols obviate need for proxies
• Builds on critical mass and momentum

8
Do they Scale?

• XML Web Services conceived to solvee-business
  interop problem
• Implementations geared towards high-end computers
• The same interop problem is the crux of
  ubiquitous computing
• Resource constraints
• Silicon footprint
• Energy parsing overhead
• Bandwidth size of messages
• ? Efficient implementation and compression

9
Specific Needs

• Coordinated real-time computing on multiple
  devices
• Energy aware protocols (turn off receiver)
• Efficient Security and Privacy
• Simple setup
• Performance

10
SOAP example
Add two numbers

• ltsoapEnvelope xmlnssoaphttp//schemas.xmlsoap.o
  rg/soap/envelope/gt
• ltsoapHeader
• soapencodingStylehttp//schemas.xmlsoap.
  org/soap/encoding/gt
• ltwsaTo xmlnswsahttp//schemas.xmlso
  ap.org/ws/2004/08/addressinggt
• http//172.31.46.26/COB/calc.cob
• lt/wsaTogt
• lt/soapHeadergt
• ltsoapBody
• soapencodingStylehttp//schemas.xmlsoap.
  org/soap/encoding/gt
• lttnsAdd xmlnstnshttp//tempuri.org/Calc/mes
  sage/gt
• ltAgt14lt/Agt
• ltBgt28lt/Bgt
• lt/tnsAddgt
• lt/soapBodygt
• lt/soapEnvelopegt 

The calculator is a popular interop test
11
Outline

• Invisible Computing Devices
• Why Web Services?
• Our approach
• Table driven serialization
• Compressed SOAP
• Trust and secure discovery
• Distributed real-time
• Componentized RTOS
• Real-time C
• Result
• Call to action

12
Table Driven Serialization

• Processes messages automatically according to
  description
• XML metadata description
• Compiled offline into compact description
• Extensible at runtime
• Process while receiving
• Zero copy networking
• Serializer parser share buffers with network
  stack crypto
• COM-Lite automation
• Turns messages into object calls
• Multiple methods in one message
• Multiple transports and encodings
• UDP, HTTP, Encryption, Compression

13
Compressed SOAP

• Goal Add(A,B) call should be three bytes
• 1) Add 2) A 3) B
• Must be convertible to XML mechanically
• Must be able to represent any XML

• Template based compressor
• Onus on discovery process
• Process compressed data directly
• Simple device only implements what it cares for

14
A Secure Invisible Home
15
Trust and Discovery

• Simple SOAP based trust and service discovery for
  ad hoc networks
• Use Big Web Services (WSE/Indigo) when
  infrastructure present
• Optimized for cluster of nodes. Base station (PC)
  deals with global issues
• Public Key Cryptography works on small devices
  (but can be boosted)
• 13s RSA decrypt, 0.03s AES on 25MHz Arm7
• FPGA takes times down by factors of 3000 and
  10000 (3ms 2µs)
• Strong crypto necessary for marketability
• WS-Security is too bulky

16
Setting up a Secure Home

• Create house authority, e.g. watch
• Touch each device once with watch
• Admits device into trust domain
• Discovery process finds device with desired
  function does key exchange
• House authority can be offline
• RSA AES
• Write hash of house authoritys key on check to
  establish trust with bank
• ? Federation of independent trust domains

17
Intuitive Setup

• Exploit Context Histories
• Trust setup touch based
• The touching history is data mined to find
  proximity clusters
• The proximity clusters indicate functional
  relationship
• E.g. touch light-switch and light close together
• automatically ties them together

18
Outline

• Invisible Computing Devices
• Why Web Services?
• Our approach
• Table driven serialization
• Compressed SOAP
• Trust and secure discovery
• Distributed real-time
• Componentized RTOS
• Real-time C
• Result
• Call to action

19
Distributed Real-Time

• Coordinated activities
• Statistical quality control

Planner
Scheduling
Producer
Scheduling
Sampling
Sensor readings
Consumer
20
Real Time...continued

• Instigator of activity plans and tunes
  reservations based on feedback samples
• Planner pre-declares future activity at each node
• Serialize scheduling trees into XML
• Worker nodes accept/reject schedules
• Coordinated schedules allow shared resource
  scheduling. Could turn off radio to save energy
• Statistical decision making
• Quality control sampling
• Concept demo shown at booth 320

21
Behavior Patterns
Define the actions and resources for a job Given
to planner

• ltbehavior name"SensorDemo"gt
• ltaction name"DemoInstigator"
  endpoint"nodeinstigator/COB/sensormain.cob"gt
• ltmessage destination"SensorProducer/"/gt
• lt/actiongt
• ltaction name"SensorProducer"
  endpoint"nodesensor/COB/sensor.cob"/gt
  ltrepeat count"100" Period"P1.5S"/gt
• ltmessage destination"SensorConsumer"/gt
• lt/actiongt
• ltaction name"SensorConsumer
• endpoint"nodeconsumer/COB/sensor.cob"/
  gt
• ltrepeat count"100" periodP1.5S/gt
• lt/actiongt
• ltsampling destination"nodeinstigator
• interval"20" number"2"/gt
• lt/behaviorgt

22
Task Instance of Pattern
Specific time and place Sent to worker by planner

• lttask name"SensorDemo-123456"gt
• ltaction name"SensorProducer
• deadline2004-12-11T025148.7001508Z
• toleranceP0.005S durationP0.02Sgt
• lttrigger maxCount"100
• offset"P1.5S"gtSensorProducerlt/trigge
  rgt
• lt/actiongt
• ltsampling destinationhttp//10.10.10.10/feedbac
  k
• interval"20" number"2"/gt
• lt/taskgt

23
Resource and Time Estimation

• Application specifies confidence needed
• Higher confidence costs more resources

estimate mean confidence
24
Action Plan

• Estimate is predicted by statistical model
• Tolerance is medium dependent

earliest start
latest start
deadline
tolerance
estimate mean confidence
25
Action Plan
Measurements are fed into statistical model Model
predicts future needs
Period k
Fixed producer deadline
1
Time t
kt0
kt2
kt1
Producer
Transmission
Consumer
26
Timing Adjustment
Measurements are fed into statistical model Model
predicts future needs
Period k
Fixed producer deadline
1
Time t
kt0
kt2
kt1
Producer
Transmission
Consumer
27
Timing Adjustment
Measurements are fed into statistical model Model
predicts future needs
Period k
Fixed producer deadline
1
Time t
kt0
kt2
kt1
Producer
Transmission
Consumer
28
Timing Adjustment
Measurements are fed into statistical model Model
predicts future needs
Period k
Fixed producer deadline
1
Time t
kt0
kt2
kt1
Producer
Transmission
Consumer
29
Timing Adjustment
Measurements are fed into statistical model Model
predicts future needs
Period k
Fixed producer deadline
1
Time t
kt0
kt2
kt1
Producer
Transmission
Consumer
30
Timing Adjustment
Measurements are fed into statistical model Model
predicts future needs
Period k
Fixed producer deadline
Steady state
1
Time t
kt0
kt2
kt1
Producer
Transmission
Consumer
31
Timing Adjustment
Measurements are fed into statistical model Model
predicts future needs
Fixed consumer deadline
Fixed producer deadline
1
1
Time t
Time t
kt0
kt2
kt1
kt0
kt1
kt2
CD player to speaker
Sensor to recorder
32
Outline

• Invisible Computing Devices
• Why Web Services?
• Our approach
• Table driven serialization
• Compressed SOAP
• Trust and secure discovery
• Distributed real-time
• Componentized RTOS
• Real-time C
• Result
• Call to action

33
Componentized RTOS

• Support for web services on a chip
• General purpose in the abstract. Code and
  interface reuse
• Special in the concrete. Only take what you need.
• Component-based objects everywhere
• Real-time scheduling with application feedback
• XML-based configuration and communication
• Hardware platforms
• ARM (many), PPC, i386, MIPS, TriMedia, Map1000,
  H8, eCOG1
• Numerous development boards, prototype gadgets,
  smart I/O cards
• Perfect research platform elusive

34
Real-Time C

• CLR desirable option for embedded systems
• Great for extensions, games, apps
• Not practical as the exclusive solution in
  embedded systems

• Our real-time scheduling extensions
• Prototype C API implemented
• Work-item Scheduler allows mixing native and
  managed threads
• Interpreter execution stacks are multiplexed
• No ROM size penalty

35
Outline

• Invisible Computing Devices
• Why Web Services?
• Our approach
• Table driven serialization
• Compressed SOAP
• Trust and secure discovery
• Distributed real-time
• Componentized RTOS
• Real-time C
• Result
• Call to action

36
Microsoft Invisible Computing

• A software platform for low cost embedded
  systems that communicate with each other and with
  big computers
• Research prototype
• Has actively been used by academia
• Community source license
• Complete TCP/IP, HTTP, SOAP, compressed SOAP,
  real-time SOAP, automation, discovery, trust
  security, RTOS (dynamic memory, threads,
  drivers, etc.), application.
• ? Runs in computer with 32KB of RAM, 256KB of
  ROM.
• Fewer components ? smaller footprint. TCP/IP is
  biggest hog. Crypto not optimized for size.

37
Call to Action

• Visit booth 320
• Download software from http//research.microsoft.
  com/invisible
• Use it in products and research
• Make your product interoperate with low-cost
  devices
• Talk to jvh